Pytorch学习笔记（I）——预训练模型（二）：修改网络结构（ResNet50及以上）

（pytorch1.0）最近在研究pytorch如何修改与训练模型的网络结构，然后发现了两种版本，一种是细调版，一种是快速版
　　经过一番钻研后发现细调版适合对网络模型进行大幅度的改动（如在原有的结构上穿插着增减层），而快速版适合直接对网络末端的层进行增减。
　　虽然快速版简单易懂，但是还是要对细调版有所了解才能比较，万一以后用的上呢。因此，我就好好研究了一番细调版，结果发现网上的代码或者博客基本都是相互搬运的，代码中的错误一模一样，对于我这种小白来说特别不友好。于是，我就在前人的基础上查缺补漏，重新整理了一下。
　　关于如何加载和使用，请查看前一篇博客Pytorch学习笔记（I）——预训练模型（一）：加载与使用

一、细调版

话不多说，直接上代码，这里以resnet50为例
这一步必须参考原来的网络结构，从而定义一个类似的网络

import torchvision.models as models
import torch
import torch.nn as nn
import math
import torch.utils.model_zoo as model_zoo

#Bottleneck是一个class 里面定义了使用1*1的卷积核进行降维跟升维的一个残差块，可以在github resnet pytorch上查看
class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,
                               padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(planes * 4)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        residual = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            residual = self.downsample(x)

        out += residual
        out = self.relu(out)

        return out

#不做修改的层不能乱取名字，否则预训练的权重参数无法传入
class CNN(nn.Module):

    def __init__(self, block, layers, num_classes=9):
        self.inplanes = 64
        super(CNN, self).__init__()
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
        self.avgpool = nn.AdaptiveAvgPool2d(output_size=(1,1))
        # 新增一个反卷积层
        self.convtranspose1 = nn.ConvTranspose2d(2048, 2048, kernel_size=3, stride=1, padding=1, output_padding=0,
                                                 groups=1, bias=False, dilation=1)
        # 新增一个最大池化层
        self.maxpool2 = nn.MaxPool2d(kernel_size=3, stride=1, padding=1)
        # 去掉原来的fc层，新增一个fclass层
        self.fclass = nn.Linear(2048, num_classes)

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2. / n))
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.inplanes, planes * block.expansion,
                          kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for i in range(1, blocks):
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)
	#这一步用以设置前向传播的顺序，可以自行调整，前提是合理
    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        # 新加层的forward
        x = x.view(x.size(0), -1)
        x = self.convtranspose1(x)
        x = self.maxpool2(x)
        x = x.view(x.size(0), -1)
        x = self.fclass(x)

        return x


# 加载model
resnet50 = models.resnet50(pretrained=True)
#3 4 6 3 分别表示layer1 2 3 4 中Bottleneck模块的数量。res101则为3 4 23 3 
cnn = CNN(Bottleneck, [3, 4, 6, 3])
# 读取参数
pretrained_dict = resnet50.state_dict()
model_dict = cnn.state_dict()
# 将pretrained_dict里不属于model_dict的键剔除掉
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
# 更新现有的model_dict
model_dict.update(pretrained_dict)
# 加载我们真正需要的state_dict
cnn.load_state_dict(model_dict)
# print(resnet50)
print(cnn)

接下来我们来比对一下前后的变化。
1、 原来的resnet50最后两层信息如下

(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)

2、 新的最后几层层信息如下

(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(convtranspose1): ConvTranspose2d(2048, 2048, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(maxpool2): MaxPool2d(kernel_size=3, stride=1, padding=1, dilation=1, ceil_mode=False)
(fclass): Linear(in_features=2048, out_features=9, bias=True)

可以看出在最后一层fc被替换了。
如果想在中间进行增减，改变前向传播的顺序就好了

二、快速版

这里以vgg16为例

import torchvision.models as models
import torch
import torch.nn as nn

class Net(nn.Module):
    def __init__(self, model):
        super(Net, self).__init__()
        # -2表示去掉model的后两层
        self.vgg_layer = nn.Sequential(*list(model.children())[:-2])
        self.transion_layer = nn.ConvTranspose2d(2048, 2048, kernel_size=14, stride=3)
        self.pool_layer = nn.MaxPool2d(32)
        self.Linear_layer = nn.Linear(2048, 8)

    def forward(self, x):
        x = self.resnet_layer(x)
        x = self.transion_layer(x)
        x = self.pool_layer(x)
        #将一个多行的Tensor,拼接成一行,-1指在不告诉函数有多少列
        x = x.view(x.size(0), -1)
        x = self.Linear_layer(x)
        return x

vgg = models.vgg16(pretrained=True)
model = Net(vgg)

1、 原来的vgg16特征提取之后有一个pooling层和分类器

(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
  (0): Linear(in_features=25088, out_features=4096, bias=True)
  (1): ReLU(inplace)
  (2): Dropout(p=0.5)
  (3): Linear(in_features=4096, out_features=4096, bias=True)
  (4): ReLU(inplace)
  (5): Dropout(p=0.5)
  (6): Linear(in_features=4096, out_features=1000, bias=True)

2、 新的最后几层层信息如下

(transion_layer): ConvTranspose2d(2048, 2048, kernel_size=(14, 14), stride=(3, 3))
(pool_layer): MaxPool2d(kernel_size=32, stride=32, padding=0, dilation=1, ceil_mode=False)
(Linear_layer): Linear(in_features=2048, out_features=8, bias=True)

我用附录的代码查看后，发现权值已经载入到新的模型了，不用像细调版那样。

附录、查看权值

#查看vgg16
for child in vgg.modules():
    print(child)
    for param in child.parameters():
        print(param)
        
 #查看新的Net
 for child in modules.modules():
    print(child)
    for param in child.parameters():
        print(param)